Please note, responding to Pieter and Chuck concurrently.

On Thu, 30 Jan 2014 07:16:54 -0800, Pieter Wuille  
<> wrote:
> Currently, with the specification and implementation in Bitcoin Core,
> if a merchant wants to use the refund or memo feature, they need to
> provide an alternative route for delivering that information to them
> *before* the transaction is made, as sending the transaction may
> result in the transfer of funds without knowing what to do with it (if
> their receive server is down at the right time) and potnetially no way
> to contact the sender. This makes these fields utterly useless.
> This is not a matter of letting wallets experiment with the best
> behaviour. This is removing the ability to rely on the payment
> protocol being bidirectional.

I think we want to separate the two issues;

   1) Reliably getting refund/memo fields to the merchant/payee
   2) Who broadcasts a TX, how it's retried, how outputs are 'locked' and  
if/when they should be [double]-spent to clear them

We should be able to solve '1' without having to fully spec out behavior  
for 2.

On Thu, 30 Jan 2014 07:16:54 -0800, Pieter Wuille  
<> wrote:
> I don't care whether wallets broadcast the transactions or not (they
> can experiment with that as they like). But we should take measures to
> prevent a transaction for being broadcast without the payment being
> delivered. One way is never broadcasting the transaction yourself.
> Another is retrying to send the payment if delivery fails.
> Here is what I would suggest to add to the specification:
> * If a payment_uri is specified, the client must attempt to send the
> payment there.
> * If a transaction is broadcast (which is permitted even if sending
> the payment fails), a client should make a reasonable attempt of
> delivering the payment (remembering, retrying, ...).
> * If a paymentACK has been received, the client is no longer
> responsible for broadcasting the transaction (but still may).

To reliably deliver refund/memo fields, we could;

   a) Send them as part of the initial request for the  
   b) Send them as a response to the PaymentRequest/PaymentDetails before  
the transaction is even formed and any unspent outputs are selected
   c) Send them as a response to the PaymentRequest/PaymentDetails with the  
UNsigned transaction, and then follow up with the signed transaction in a  
separate message.

'a' is problematic because while wallet software could easily append some  
data to the queryString, it doesn't work if the user is downloading then  
opening the PaymentRequest as a file. So 'a' is a no-go I think.

'b' is fine, if not overly chatty. The only thing committed is a refund  
address, which is a lot less problematic than committed unspent outputs.

'c' is nice because it lets the server preview the transaction (and  
ACK/NACK it with a memo of their own -- e.g. 'fee too low'?) without being  
able to broadcast it, so we know unspent outputs are not yet committed.

But all of these require too many changes to the protocol for my liking.

On Wed, 29 Jan 2014 21:47:51 -0800, Chuck <>  
> 3. Customer builds a set of transactions and sends a new 
> PaymentApprovalRequest message which includes a refund address and the 
> unsigned transactions and their associated fully-signed transactionhash,  
> the whole message signed with the private key of the refund address.

"Unsigned transactions and their associated fully-signed transaction hash"  
-- isn't that a fully signed transaction? In this case, it doesn't solve  
the core problem of the server being able to broadcast that transaction  
without ACKing.

On Wed, 29 Jan 2014 21:47:51 -0800, Chuck <>  
> In Step 3, it's critical the customer sign the message with the private
> key of the refund address, so that the merchant can be confident the
> refund address is correct.

For merchant confidence that the address is correct, we can leave the  
transport security to the transport layer.

For payer proving refund address was X after merchant sends a refund to Y,  
that's a different story. I don't think we want to *require* access to the  
refund address private key. For example, BIP32 public derivation may have  
just the pubkey available. Offline transaction signing is one thing, but  
offline PP message signing is too much. I think there are better ways to  
secure the refund address which can reuse existing code, and certainly the  
option should remain to send a plain refund address just relying on  
transport security and trusting the merchant.

WatchGuard Dimension instantly turns raw network data into actionable 
security intelligence. It gives you real-time visual feedback on key
security issues and trends.  Skip the complicated setup - simply import
a virtual appliance and go from zero to informed in seconds.
Bitcoin-development mailing list

Reply via email to